Thursday, January 29, 2009

Optimal Optimizing Program

create a program that reads a program in machine language and does some periphery optimizations on it wherever it can, even if they're a little expensive

run the program on itself, then modify the program slightly so that, with its improved speed, it can afford to make some more serious optimizations

run the program on itself again, repeat cycle until your program completely refactors machine code to make it better.

the ideal goal would be to make a program that takes a program as input, and gives the most optimal possible program that does the exact same thing. that, per se, may not be practical, but it would be nice to see how far we can go with the idea.

once we have a pretty good optimizing program, we can run it once for a given program, even on a supercomputer if necessary, and then disseminate the result to all the end users.

there probably is not an optimal optimizing program that creates the best possible optimization in earthly time, because even if it had come up with clever ways to make specific kinds of optimizations, it could never *prove* that the result is the best possible optimization, in the most general case, without iterating through every possible program to verify. so a program that does *perfect*, per se, optimization, may not be a useful thing to aim for, that is by trying to optimize a program that originally *does* iterate through every possibility.
that same idea applies if we were to make the best sound compression software, etc.

Wednesday, January 14, 2009

Heisenberg Uncertainty

Consider a sympathetic resonator from which you take intensity samples 30 times every second. Now consider that its resonant frequency is 20 hertz. So at each sample you may get an exact number, which depends on all its previous activity, but yet, since a single wave itself is larger than the sampling period, the exact number doesn't have complete meaning. Or the time you sample it at doesn't have complete meaning. Either way you look at it.

This is reminiscent of the Heisenberg uncertainty principle, by which two variables like position or velocity can be individually known but their combined precision (multiplied) can't exceed h/2, and also since it's about waves it alludes to wave-particle duality.

So the two ways of looking at the amplitude/phase problem in the resonator example could correspond, somehow, to the two ways of sampling an object: position, or velocity. A different way of sampling it is a different way of looking at the object.

Saturday, January 10, 2009

The next level of HTTP tags

inhahe (6:12:28 PM): inhahe (5:46:22 PM): the idea is a simple mechanism to highlight a particular part of a webpage, which adds something to the url string
inhahe (5:46:41 PM): which you can then send to another person so they can see the webpage with the same thing highlighted, circled, whatever.
inhahe (5:47:01 PM): it would be an extension to the http syntax
inhahe (5:47:08 PM): /writes it
inhahe (5:47:32 PM): i see problems with the idea
inhahe (5:47:37 PM): hmm
inhahe (5:48:28 PM): nah i guess it works
phaldo (5:48:47 PM): why not just copy the text you want to highlight
inhahe (5:49:54 PM): that too but there are several reasonsyou might not want to. 1, if you copy the text, the formatting, font sizes colors, graphics etc. might not cross over depenidng on the medium your'e using to chat with
inhahe (5:50:01 PM): 2. a url is a lot shorter
inhahe (5:50:15 PM): 3. it's more useful to use as citations or references
inhahe (5:50:53 PM): i think the only problem with the idea is that if a document changes, the highlight information could become wrong or invalid
inhahe (5:50:58 PM): i just reaolized that
inhahe (5:51:07 PM): but we can just say too bad and let it become wrong or invalid in those cases